outcomerate is a lightweight R package that implements
the standard outcome rates for surveys, as defined in the Standard
Definitions, 10th edition of the American Association for Public
Opinion Research (AAPOR).
Although the mathematical formulas are straightforward, it can get
tedious and repetitive calculating all the rates by hand, especially for
sub-groups of your study. The formulas are similar to one another and so
it is also dangerously easy to make a clerical mistake. The
outcomerate package simplifies the analytical workflow by
defining all formulas as a collection of functions.
The 10th edition separates code 3.20 from UO under the
aggregate symbol UR. Legacy UO data remain
supported and produce the same rates when the package’s scalar
eligibility estimate e is used; newly coded 3.20 cases
should use UR.
Install the package from CRAN:
install.packages("outcomerate")Alternatively, install the latest development version via github:
#install.packages("devtools")
devtools::install_github("ropensci/outcomerate")Let’s say you draw a sample of 13 cases. After finishing the fieldwork, you tabulate all your attempts into a table of disposition outcomes:
| code | disposition | n |
|---|---|---|
| I | Complete interview | 4 |
| P | Partial interview | 2 |
| R | Refusal and break-off | 1 |
| NC | Non-contact | 1 |
| O | Other | 1 |
| UH | Unknown if household | 1 |
| UR | Unknown if sampled unit is eligible / housing unit contains an eligible respondent | 1 |
| NE | Not eligible | 1 |
| UO | Unknown, other | 1 |
Using this table, you may wish to report some of the common survey outcome rates, such as:
Most of these rates come under a number of variants, having
definitions that are standardized by AAPOR. The outcomerate
function lets you calculate these rates seamlessly:
# load package
library(outcomerate)
# set counts per disposition code (needs to be a named vector)
freq <- c(I = 4, P = 2, R = 1, NC = 1, O = 1,
UH = 1, UR = 1, UO = 1, NE = 1)
# calculate rates, assuming 90% of unknown cases are eligible
outcomerate(freq, e = eligibility_rate(freq))
#> RR1 RR2 RR3 RR4 RR5 RR6 COOP1 COOP2 COOP3 COOP4 REF1 REF2 REF3
#> 0.333 0.500 0.342 0.513 0.444 0.667 0.500 0.750 0.571 0.857 0.083 0.085 0.111
#> CON1 CON2 CON3 LOC1 LOC2
#> 0.667 0.684 0.889 0.750 0.769When the available evidence supports different eligibility estimates for the unknown categories, pass them as a named vector. Each value is the probability that a case in that category is eligible:
e_by_class <- c(UH = 0.4, UR = 0.7, UO = 0.2)
outcomerate(freq, e = e_by_class, rate = c("RR3", "REF2", "CON2"))
#> RR3 REF2 CON2
#> 0.388 0.097 0.777A category may be omitted from a non-scalar e only when
its aggregate count is zero (after weighting, when weights are
supplied). A length-one value—including the result of
eligibility_rate()—keeps the original behavior and applies
to every unknown category.
Dispositions do not always come in a tabulated format. Survey
analysts often work with microdata directly, where each row represents a
sampled case. The outcomerate package allows you to obtain
rates using such a format as well:
# define a vector of dispositions
x <- c("I", "P", "I", "UO", "R", "I", "NC", "I", "O", "P", "UH", "UR")
# calculate desired rates
outcomerate(x, rate = c("RR2", "CON1"))
#> RR2 CON1
#> 0.50 0.67
# obtain a weighted rate using illustrative base weights
w <- c(rep(1.3, 6), rep(2.5, 6))
outcomerate(x, weight = w, rate = c("RR2", "CON1"))
#> RR2w CON1w
#> 0.45 0.61